-
Notifications
You must be signed in to change notification settings - Fork 0
Voice setup
Lwando Zwane edited this page Aug 18, 2026
·
1 revision
Voice is fully local — audio never leaves your machine.
-
Hands-free — click the 🎤 button in the top bar (or
⌘⇧V). The user talks, the assistant answers aloud, and no pane opens. The transcript is the assistant's. - Push-to-talk — the 🎙 button in the chat composer records while you hold (or click to toggle); the transcript lands in the composer for you to edit before sending.
Tome transcribes with one of two engines, selectable in Preferences → Voice:
| Engine | Setup | Notes |
|---|---|---|
| Apple on-device (default, macOS) | None | Streaming, multilingual, zero install |
| whisper.cpp (fallback / Linux / offline) |
brew install whisper-cpp + model |
Fully local; batch transcription |
- Auto uses Apple when available, otherwise whisper.
- With whisper selected, install the binary once (
brew install whisper-cpp), then click Download speech model in Preferences → Voice (or pointTOME_WHISPER_BINat a custom binary).
- Barge-in — start talking while the assistant is speaking (or thinking) to interrupt; toggle in the voice button's right-click menu.
- Push-to-talk mode — dictate into the composer without auto-sending.
- Voice — pick the text-to-speech voice in the same menu.
- Warm up whisper at launch — loads the model in the background so the first dictation is instant.
See Troubleshooting for the common "whisper-cli not found" and microphone errors.
Tome on GitHub · Report issues in the repository, not the wiki.